If robots are to be actively used in production, they must learn to predict human movements in order to be effective. For example, understanding that a particular tool is required to perform a particular action and providing it in a timely manner. Know the assembly sequence and transfer the next part.

what is known

Computer science researchers from the University of Southern California in Viterbo (USA) have found a way to teach a robot to predict how a given person is more likely to solve a task. The authors focused on educational robots that would “hand out the keys,” that is, assist a person during repairs, assembly, and even table setting.

Usually, people choose one of two strategies when solving such problems: some start from the most difficult, others – from the simplest. At the same time, the habit, as a rule, does not change throughout life.

Until now, in order to teach a robot the sequence of operations, it was necessary to show it the whole operation from start to finish. It takes a lot of time and effort – such an “assistant” loses its meaning. in a new study the robot was taught to determine which strategy a person would follow in various tasks.

For this, a simple airplane model is added to the robot, which is recommended to be mounted in front of the camera of the machine. AI will observe and draw conclusions. And then it remains only to load the instructions for the specific object that is planned to be assembled, and the robot provides the parts and tools in the necessary order. During testing, the system was able to “predict” the sequence of people’s actions with about 82% accuracy.

Experiment recording: video

A robot that quickly understands your preferences will help prepare meals, rearrange furniture or make repairs, and will have a significant impact on our daily lives.
– says one of the developers Stefanos Nikolaidis.

Many applications can be invented for such machines, constantly improving their skills. However, it is unclear when this technology will hit the market.